| Targeted News Service |
Most actuaries know that predictive modeling — the harnessing of enormous data sets to do everything from rate baseball players to price auto insurance — is the hot trend among the math-literate. Most CEOs know it, too. But many hesitate to embrace it — daunted perhaps by how long a model can take to get launched or how much it will cost.
How can actuaries convince executives to take the plunge, sooner rather than later? A panel of high-ranking actuaries and executives offered ideas at a presentation called "What Executives Need to Know about Predictive Modeling," part of the
First, some history: Predictive analytics has changed the property/casualty insurance industry in the past decade and a half, though a bit slower than it could have been, according to
"Property and casualty is not the last industry to discover analytics, but it was somewhat slower than others," he said.
During his remarks, Ellingsworth emphasized the key to accelerated adoption was coalescing analytic expertise in an industry with many resources and disparate approaches to operational efficiency. He also explained how actuarial professionals and predictive modelers are increasingly aligned with business leaders to dramatically improve many facets of insurance operations in marketing, risk assessment, rating, underwriting, claims, agent/customer service, catastrophe analytics, and capital risk management. These efforts now help to reduce the uncertainty around carrier financial performance, he said.
The old analysis clearly did not work. So, insurance executives began using complex computer models to determine their exposures.
The models, Mildenhall said, "gave us a ruler to measure risk in a consistent way."
"They said, 'Why aren't you doing more?'"
At her company, "The case was made for us by [Hurricanes] Hugo and Andrew after our existing models, like those at many other companies, did not perform adequately." So the catastrophe models made the case for themselves.
Other situations to advocate for models prove more difficult, Gannon said. She advised:
* Pitch the model at a high-level. "Start at the 50,000-foot level and don't plunge into the nitty-gritty too fast."
* Estimate the return on investment. Show how your project aligns with the executives' other objectives.
* Emphasize data quality. "You've got to make sure the data's good."
* Be clear about the goal of the model. "Make sure this thing is going to move the needle you specify."
* Build on prior successes. "The executive will listen to the previously successful modeler more than the rookie. If you're the rookie, work with someone else to get your shot."
* Put together a strong modeling team, she said, adding that actuaries have a strong skill set for a modeling team. And be sure to invest to sustain your intellectual capital.
* But at the same time be clear about the risks that exist in building the model.
ISO's Ellingsworth said one should emphasize that modeling is a process, not a single project. It's a "cultural statement that you are in it, and are in it to win it," he said.
He added that executives are changing with the times and starting to expect deep dives into data. "Executives are becoming infoholics," he said. If you can show them something they didn't know and show how the company benefits from knowing it, "be ready to be challenged how fast you can do that."
And you need to monitor the investment in the model, he said. If it's working well, the executive team will want to increase that investment.
In addition to the overall return on investment, a pitch for a new model should show how it will affect retention ratios, as well as address issues like impact on the regulatory environment, consumers, and the company's distribution channels.
He also suggested telling the executives how the project will be measured and giving them an idea of what reports it will generate. "Give examples of how the model will help the company do things faster, better and cheaper," Bauer explained.
"You have to model the model," he said, "and show how it performs under a wide variety of conditions, including scenarios selected completely at random."
A company might also consider how its models will affect competitors. Progressive was one of the first companies to put a rating engine online. It not only showed Progressive's rates, but those of competitors. It helped the company for several reasons.
For one, it got traffic, which built awareness of Progressive's name. Showing competitor rates conveyed the sense that the company was an honest broker. Sometimes, of course, it sent business to a competitor. But that was OK, too, Bauer said. Suppose the rating engine showed that a competitor prices a policy at
"If we did our pricing right," Bauer said, "we just sent our competitor a
That, of course, was more than a decade ago. Now predictive models have 'big data' — billions and billions of data points that can let a company pinpoint customer preferences — if it only can be tamed.
In such cases, too much information can be the problem, Gannon said. She worked on a project that tried to model off 795 variables. The first step was to narrow that down to 150.
But it's important to take on such large projects, she said. Customers increasingly want a customized solution. If you provide that, or convince customers they want something better, your company can be successful.
Bauer predicted that 'big data' will allow companies to go beyond today's predictive models to more advanced cluster pricing, which will provide for more refined expense pricing.
The panelists agreed that companies will see increasing advocates for models and strong analytics, and for good reason.
"The winners for the future are the ones who are building analytic models," Gannon concluded.
TNS rd43 121214-JF78-4140665 StaffFurigay
| Copyright: | (c) 2012 Targeted News Service |
| Wordcount: | 1178 |
More Articles